1 files changed, 310 insertions, 0 deletions

diff --git a/src/ot_socket.c b/src/ot_socket.c
new file mode 100644
index 0000000..79bb01a
--- /dev/null
+++ b/src/ot_socket.c
@@ -0,0 +1,310 @@
+/* This software was written by Dirk Engling <erdgeist@erdgeist.org>
+   It is considered beerware. Prost. Skol. Cheers or whatever. */
+
+/* System */
+#define _GNU_SOURCE /* accept4 on glibc */
+#include <errno.h>
+#include <fcntl.h>
+#include <netinet/in.h>
+#include <netinet/tcp.h>
+#include <string.h>
+#include <sys/socket.h>
+#include <unistd.h>
+
+/* Opentracker */
+#include "ot_ip.h"
+#include "ot_socket.h"
+
+static void sockaddr_from_ip(struct sockaddr_in6 *sa, const ot_ip6 ip, uint16_t port) {
+  memset(sa, 0, sizeof(*sa));
+  sa->sin6_family = AF_INET6;
+  sa->sin6_port   = htons(port);
+  memcpy(&sa->sin6_addr, ip, sizeof(ot_ip6));
+}
+
+/* Linux reports an IPv4 peer on a dual stack socket as a v4 mapped
+   sockaddr_in6, the BSDs and macOS hand back a plain sockaddr_in instead, so
+   both forms have to be read. */
+static void ip_from_sockaddr(ot_ip6 ip, uint16_t *port, const struct sockaddr_storage *ss) {
+  if (ss->ss_family == AF_INET) {
+    const struct sockaddr_in *sa = (const struct sockaddr_in *)ss;
+
+    memcpy(ip, ip_v4mapped_prefix, sizeof(ip_v4mapped_prefix));
+    memcpy(ip + sizeof(ip_v4mapped_prefix), &sa->sin_addr, 4);
+    *port = ntohs(sa->sin_port);
+  } else {
+    const struct sockaddr_in6 *sa = (const struct sockaddr_in6 *)ss;
+
+    memcpy(ip, &sa->sin6_addr, sizeof(ot_ip6));
+    *port = ntohs(sa->sin6_port);
+  }
+}
+
+int socket_set_blocking(int s, int blocking) {
+  int flags = fcntl(s, F_GETFL, 0);
+
+  if (flags == -1)
+    return -1;
+  return fcntl(s, F_SETFL, blocking ? (flags & ~O_NONBLOCK) : (flags | O_NONBLOCK));
+}
+
+static int set_nonblocking(int s) { return socket_set_blocking(s, 0); }
+
+/* Close without letting the failure we are about to report be overwritten */
+static int close_and_fail(int s) {
+  int saved = errno;
+  close(s);
+  errno = saved;
+  return -1;
+}
+
+static int socket_v6(int type) {
+  int s = socket(PF_INET6, type, 0);
+  int zero = 0;
+
+  if (s == -1)
+    return -1;
+
+  /* A tracker that answered only one family would be half a tracker */
+  if (setsockopt(s, IPPROTO_IPV6, IPV6_V6ONLY, &zero, sizeof(zero)) == -1)
+    return close_and_fail(s);
+
+  if (set_nonblocking(s) == -1)
+    return close_and_fail(s);
+
+  return s;
+}
+
+/* SO_REUSEPORT means three different things. Linux spreads datagrams over
+   the group. FreeBSD permits the bind and spreads nothing, which is what
+   SO_REUSEPORT_LB was added for. Darwin hands every datagram to whichever
+   socket bound last, so a group there starves all but one reader. */
+#if defined(SO_REUSEPORT_LB)
+#define OT_BALANCE_OPTION SO_REUSEPORT_LB
+#elif defined(SO_REUSEPORT) && !defined(__APPLE__)
+#define OT_BALANCE_OPTION SO_REUSEPORT
+#endif
+
+int socket_can_balance(void) {
+#ifdef OT_BALANCE_OPTION
+  return 1;
+#else
+  return 0;
+#endif
+}
+
+/* The only place in the tree that needs to know what a kernel calls this */
+static void socket_set_sharing(int s, socket_sharing sharing) {
+  int one = 1;
+
+  setsockopt(s, SOL_SOCKET, SO_REUSEADDR, &one, sizeof(one));
+#ifdef OT_BALANCE_OPTION
+  if (sharing == SOCKET_BALANCED) {
+    setsockopt(s, SOL_SOCKET, OT_BALANCE_OPTION, &one, sizeof(one));
+    return;
+  }
+#endif
+#ifdef SO_REUSEPORT
+  if (sharing != SOCKET_EXCLUSIVE)
+    setsockopt(s, SOL_SOCKET, SO_REUSEPORT, &one, sizeof(one));
+#endif
+}
+
+static int socket_bind_v6(int s, const ot_ip6 ip, uint16_t port, socket_sharing sharing) {
+  struct sockaddr_in6 sa;
+
+  socket_set_sharing(s, sharing);
+
+  sockaddr_from_ip(&sa, ip, port);
+  return bind(s, (struct sockaddr *)&sa, sizeof(sa));
+}
+
+/* Defer the listening socket reporting it to be ready to accept until enough
+   sensible data is available for an accept+read pair. Where available. */
+int socket_defer_accept(int s) {
+#if defined(TCP_DEFER_ACCEPT)
+  int patience = OT_CLIENT_TIMEOUT;
+
+  return setsockopt(s, IPPROTO_TCP, TCP_DEFER_ACCEPT, &patience, sizeof(patience));
+#elif defined(SO_ACCEPTFILTER)
+  struct accept_filter_arg filter;
+  const char               name[sizeof(filter.af_name)] = "dataready";
+
+  memset(&filter, 0, sizeof(filter));
+  memcpy(filter.af_name, name, sizeof(name));
+  return setsockopt(s, SOL_SOCKET, SO_ACCEPTFILTER, &filter, sizeof(filter));
+#else
+  (void)s;
+  return -1;
+#endif
+}
+
+int socket_tcp_listen(const ot_ip6 ip, uint16_t port, socket_sharing sharing) {
+  int s = socket_v6(SOCK_STREAM);
+
+  if (s == -1)
+    return -1;
+  if (socket_bind_v6(s, ip, port, sharing) == -1 || listen(s, SOMAXCONN) == -1)
+    return close_and_fail(s);
+
+  return s;
+}
+
+int socket_udp_bind(const ot_ip6 ip, uint16_t port, socket_sharing sharing) {
+  int s = socket_v6(SOCK_DGRAM);
+
+  if (s == -1)
+    return -1;
+  if (socket_bind_v6(s, ip, port, sharing) == -1)
+    return close_and_fail(s);
+
+  return s;
+}
+
+int socket_tcp_connect(const ot_ip6 srcip, uint16_t srcport, const ot_ip6 dstip, uint16_t dstport) {
+  struct sockaddr_in6 sa;
+  int                 s = socket_v6(SOCK_STREAM);
+
+  if (s == -1)
+    return -1;
+  if (socket_bind_v6(s, srcip, srcport, SOCKET_EXCLUSIVE) == -1)
+    return close_and_fail(s);
+
+  sockaddr_from_ip(&sa, dstip, dstport);
+  if (connect(s, (struct sockaddr *)&sa, sizeof(sa)) == -1 && errno != EINPROGRESS && errno != EWOULDBLOCK)
+    return close_and_fail(s);
+
+  return s;
+}
+
+int socket_is_connected(int s) {
+  int       error = 0;
+  socklen_t len   = sizeof(error);
+
+  if (getsockopt(s, SOL_SOCKET, SO_ERROR, &error, &len) == -1)
+    return 0;
+  if (error) {
+    errno = error;
+    return 0;
+  }
+  return 1;
+}
+
+/* Linux hands over a blocking socket whatever the listening socket was, the
+   BSDs and macOS inherit the listener's flags, so the accepted socket has to
+   be told either way. accept4 carries the flag itself, which saves the two
+   fcntl calls it otherwise takes. Darwin has neither accept4 nor the flag. */
+int socket_accept(int s, ot_ip6 ip, uint16_t *port) {
+  struct sockaddr_storage ss;
+  socklen_t               len = sizeof(ss);
+  int                     fd;
+
+  memset(&ss, 0, sizeof(ss));
+#ifdef SOCK_NONBLOCK
+  if ((fd = accept4(s, (struct sockaddr *)&ss, &len, SOCK_NONBLOCK)) == -1)
+    return -1;
+#else
+  if ((fd = accept(s, (struct sockaddr *)&ss, &len)) == -1)
+    return -1;
+  if (set_nonblocking(fd) == -1)
+    return close_and_fail(fd);
+#endif
+
+  ip_from_sockaddr(ip, port, &ss);
+  return fd;
+}
+
+ssize_t socket_recv(int s, void *buf, size_t len, ot_ip6 ip, uint16_t *port) {
+  struct sockaddr_storage ss;
+  socklen_t               sslen = sizeof(ss);
+  ssize_t                 received;
+
+  memset(&ss, 0, sizeof(ss));
+  received = recvfrom(s, buf, len, 0, (struct sockaddr *)&ss, &sslen);
+  if (received == -1)
+    return -1;
+
+  ip_from_sockaddr(ip, port, &ss);
+  return received;
+}
+
+ssize_t socket_send(int s, const void *buf, size_t len, const ot_ip6 ip, uint16_t port) {
+  struct sockaddr_in6 sa;
+
+  sockaddr_from_ip(&sa, ip, port);
+  return sendto(s, buf, len, 0, (struct sockaddr *)&sa, sizeof(sa));
+}
+
+#ifdef WANT_SYNC_LIVE
+static void sockaddr4_from_ip(struct sockaddr_in *sa, const ot_ip6 ip, uint16_t port) {
+  memset(sa, 0, sizeof(*sa));
+  sa->sin_family = AF_INET;
+  sa->sin_port   = htons(port);
+  memcpy(&sa->sin_addr, ip + sizeof(ip_v4mapped_prefix), 4);
+}
+
+/* A multicast group is meant to have several listeners, so these are always
+   shared */
+static int socket_v4_bind(const ot_ip6 ip, uint16_t port) {
+  struct sockaddr_in sa;
+  int                s;
+
+  if ((s = socket(PF_INET, SOCK_DGRAM, 0)) == -1)
+    return -1;
+
+  socket_set_sharing(s, SOCKET_SHARED);
+
+  sockaddr4_from_ip(&sa, ip, port);
+  if (bind(s, (struct sockaddr *)&sa, sizeof(sa)) == -1)
+    return close_and_fail(s);
+
+  return s;
+}
+
+int socket_mcast_join(const ot_ip6 group, const ot_ip6 interface, uint16_t port) {
+  struct ip_mreq mreq;
+  ot_ip6         any;
+  int            s;
+
+  /* The group is what is listened to, so the socket takes the wildcard and
+     the join names the interface to receive it on */
+  memcpy(any, ip_v4mapped_prefix, sizeof(ip_v4mapped_prefix));
+  memset(any + sizeof(ip_v4mapped_prefix), 0, 4);
+
+  if ((s = socket_v4_bind(any, port)) == -1)
+    return -1;
+
+  memset(&mreq, 0, sizeof(mreq));
+  memcpy(&mreq.imr_multiaddr, group + sizeof(ip_v4mapped_prefix), 4);
+  memcpy(&mreq.imr_interface, interface + sizeof(ip_v4mapped_prefix), 4);
+  if (setsockopt(s, IPPROTO_IP, IP_ADD_MEMBERSHIP, &mreq, sizeof(mreq)) == -1)
+    return close_and_fail(s);
+
+  return s;
+}
+
+int socket_mcast_bind(const ot_ip6 interface, uint16_t port, int loopback) {
+  /* The BSDs and macOS want a byte for the IPv4 multicast options, Linux
+     accepts either that or an int. */
+  unsigned char ttl = 1, loop = (unsigned char)(loopback != 0);
+  int           s   = socket_v4_bind(interface, port);
+
+  if (s == -1)
+    return -1;
+  if (set_nonblocking(s) == -1)
+    return close_and_fail(s);
+
+  if (setsockopt(s, IPPROTO_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)) == -1 ||
+      setsockopt(s, IPPROTO_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)) == -1)
+    return close_and_fail(s);
+
+  return s;
+}
+
+ssize_t socket_mcast_send(int s, const void *buf, size_t len, const ot_ip6 group, uint16_t port) {
+  struct sockaddr_in sa;
+
+  sockaddr4_from_ip(&sa, group, port);
+  return sendto(s, buf, len, 0, (struct sockaddr *)&sa, sizeof(sa));
+}
+#endif